Written by: Penny Crosman
Originally published on: American Banker

For many institutions, a chatbot that uses artificial intelligence to answer customers’ questions may be seen as a bonus that helps cut costs or slightly improves the customer service.

But for Coenraad Jonker, CEO of TymeBank, the chatbot is essential to his business. With just 120 employees, his challenger bank competes with institutions that have 15,000 employers or more. Without the chatbot, that wouldn’t be possible.

“Deploying these technologies for us is not a luxury. It is a critical part of the long-term strategy,” he said. “If we want to make banking affordable, particularly for underserved segments, we have to find a way to drive the cost down. We believe to be truly affordable for our customers, we need to have less than two hundred employees at scale.”

“Deploying these technologies for us is not a luxury. It is a critical part of the long-term strategy”. – Coenraad Jonker, CEO of TymeBank

“What we are all trying to learn is what are the circumstances in which customers would get irritated by a machine and really want to speak to a human,” said Coenraad Jonker, CEO of TymeBank.

Moving customer service conversations to automated chatbots is a growing trend in the U.S. as well as the rest of the world. Six million people use Bank of America’s Erica virtual assistant. Regions Bank is using IBM Watson to answer, so far, about 22% of customers’ questions. DBS Digibank uses Kasisto’s Kai AI chatbot to handle more than 80% of customer interactions in India.

But TymeBank is unusual in the degree to which its business model relies on the technology.

High on AI

TymeBank is an international digital banking operation for emerging markets that serves underserved consumers and small businesses. Its first deployment is in South Africa, where it started with the installation of a kiosk in a supermarket at the end of February. It now has 730 kiosks in supermarkets and has opened more than 250,000 customer accounts.

“That’s a number we are very delighted with, in a country of 50 million people,” Jonker said. “That’s well ahead of our expectations.”

The first place TymeBank is applying AI is in risk management, especially credit risk. The second is in conversational banking, to “re-humanize” banking by integrating with the channels customers use in their daily lives.

“Our challenge with financial inclusion is that we are dealing with customers that tend not to want to use banking in the way that banking is offered,” Jonker said. “They do want to use banking in a way that’s more user friendly. But what we can’t afford to do is to throw armies of people at it.”

The bank is using an AI-powered chatbot from Finn AI to answer customers’ simplest questions. “Humans only engage when the chatbot runs out of things to say,” Jonker said.

Initially that means the chatbox deals with a relatively small percentage of the engagements, about 10% to 20%.

“Our job over time is to move that up to 90% and still give customers the sense that they are engaging in a human way,” Jonker said. “The idea is to move away from a world where humans have to adjust to technology and into a world where technology adjusts to humans.”

Challenges of understanding customers’ questions

Some U.S. bankers worry that a chatbot could misunderstand a customer’s question — due to differences in accent, wording and intent — and give wrong information. In South Africa, this is an even bigger issue: there are 11 official languages and English is a second language for most people.

“The learning curve is longer and steeper than it would be in environments where people speak a more standardized dialect or more standardized language,” Jonker said. “You’re never going to solve in the short term for all that variation. And the margin for error in financial services and banking in particular is lower than it is for general use chatbots. The key is to always have a human on hand.”

If TymeBank’s chatbot has any doubt or uncertainty about the meaning of a question, it refers the customer to a person.

“For delicate items, we adjust our model so that it makes sure it’s really right or it hands off to a human,” said Finn AI CEO Jake Tyler.

If someone is paying a bill, the Finn AI engine is set to have a high degree of confidence in name matching before letting that transaction through.

“That’s one of the nuances with banking is that sometimes you really have to be right and sometimes with other conversations you have a little bit more leeway from the customer’s end.

And the chatbot is designed to pick up on sensitive issues. For instance, if someone has had a death in the family or for some other reason seems emotional or frustrated, the chatbot will pass the customer to a human.

“What we are all trying to learn is what are the circumstances in which customers would get irritated by a machine and really want to speak to a human,” Jonker said. “A computer glitch is a good example where people will potentially be aggravated by getting a stock standard answer and want to hear a human on the other side.”

However, there are also times when people would rather speak to a machine than to a human.

“When I fall behind on my payments and I just want to get that payment postponed, I probably would rather speak to a chatbot, because it’s less embarrassing than speaking to another human being and saying, listen, you know, things have gone wrong and I need another five days,” Jonker said.

Finn AI has pre-mapped many of the questions retail banking customers ask frequently. It aggregates learning across the different customers it works with (including ATB Financial and Bank of Montreal in Canada and Fidor in Europe). Then it localizes the chatbots for local dialects and language as well as for customers’ specific products, according to Tyler.

TymeBank has been piloting Finn AI’s technology for a year as a virtual assistant called Max that acts as a financial fitness coach. Max guides customers to improve their credit score and save more. Max is available within the TymeCoach Android and iOS Apps as well as in Facebook Messenger and Facebook Messenger Lite. The bank is also building Finn AI into customer service and customer onboarding through its own channels as well as Facebook Messenger and WhatsApp.

“One of the things that’s interesting about the South African market and emerging markets more generally is being able to bank within third-party messaging platforms,” Tyler noted. “In South Africa, 90% of people who have access to the internet use WhatsApp. If we can deliver banking within that sort of interface that’s on people’s phones that they’re using every day, it is a very powerful channel for us to be in.”

Jonker also believes having people interact with a chatbot will keep a conversation going.

“One of the big problems for neobanks or digital banks is that they onboard customers but they don’t keep those customers engaged because, for instance, we don’t have a marketing budget that would rival our direct competitors in conventional banks,” he said.

A large part of TymeBank’s work is learning about “what we think is an inevitable part of the future model of banking,” Jonker said. “The question is not whether we going to be using AI extensively to do customer service, customer onboarding and a range of other use cases. The question is just who gets there first in the most skillful way with the least side effects or the least mistakes along the way.”

Read the case study on how TymeBank worked with Finn AI to capture an emerging market.